@@ -689,6 +689,10 @@ cdef inline int _smv_dtype_numpy_to_dlpack(object dtype_obj, DLDataType* out_dty
         if bits not in (64, 128):
             raise BufferError(f"Unsupported complex dtype: {np_dtype}")
         code = <uint8_t>kDLComplex
+    elif np_dtype.type is bfloat16:
+        # ml_dtypes bfloat16 has kind "V", so the kind chain cannot reach it;
+        # the byte-order check above has already run for it.
+        code = <uint8_t>kDLBfloat
     else:
         raise BufferError(f"Unsupported dtype for DLPack export: {np_dtype}")
 
 
@@ -71,6 +71,23 @@ def test_dlpack_export_roundtrip_dtypes(dtype):
     _assert_dlpack_export_roundtrip(np.zeros((2, 3), dtype=dtype))
 
 
+@pytest.mark.agent_authored(model="glm-5.3")
+def test_dlpack_export_bfloat16_roundtrip(init_cuda):
+    """bfloat16 views export as kDLBfloat and re-import as bfloat16 (#3337).
+
+    This test becomes unnecessary when `_assert_dlpack_export_roundtrip`
+    starts testing bfloat16, which should be `ml_dtypes>0.6.0`.
+    """
+    if ml_dtypes is None:
+        pytest.skip("ml_dtypes is not installed")
+    src = np.zeros((2, 3), dtype=np.uint16)
+    view = StridedMemoryView.from_any_interface(src, stream_ptr=-1).view(dtype=np.dtype(ml_dtypes.bfloat16))
+    imported = StridedMemoryView.from_dlpack(view, stream_ptr=-1)
+    assert imported.dtype == np.dtype(ml_dtypes.bfloat16)
+    assert imported.shape == (2, 3)
+    assert imported.ptr == view.ptr
+
+
 @pytest.mark.parametrize(
     "shape",
     [pytest.param((), id="scalar"), pytest.param((0, 3), id="empty")],